As noted by Coudouel et al. (2002; see also Ravallion, 1994), to measure poverty, one needs: a) an indicator of well-being or welfare such as per capita caloric intake or per capita expenditure; b) a threshold (the poverty line) to which each individual or household’s welfare can be compared; and c) a poverty measure. Differences in poverty estimates can result from differences in the choice of the indicator, the threshold, or the poverty measure. The main poverty measures used in empirical work have been presented in Annex 1. This annex provides information on the construction of the consumption aggregate and the poverty lines used for poverty measurement in Ghana. The annex is adapted and expanded from the material included in the poverty profile produced by the Ghana Statistical Service (2007). In addition, the annex provides estimates of poverty together with their standard errors.
Data sources
The consumption-based poverty measures presented in this study are estimated using the third, fourth and fifth rounds of the Ghana Living Standards Survey (GLSS). The GLSS is a nationally representative multi-purpose survey of households in Ghana, which collects information on many different dimensions of well-being including education, health and employment. Five rounds of data have been collected, starting in 1987/88. In this study we focus on the three most recent rounds—those conducted in 1991/92, 1998/99 and 2005/06. The questionnaires used for these three rounds were almost identical. Hence, the consumption aggregates are comparable. These total consumption of each household includes both food and non-food items (including housing). Food and non-food consumption commodities may be explicitly purchased by households, or acquired through other means (e.g. as output of own production activities, payment for work done in the form of commodities, or from transfers from other households). The household consumption takes account of all of these sources.
Construction of the consumption aggregate
The indicator of well-being used to measure poverty is the total household consumption per equivalent adult expressed in constant prices of Accra in January 2006. The first step in constructing this measure is to estimate total household consumption expenditure. Table A2.1 sets out in detail how this is done, covering the components of this, their composition and sources within the different GLSS questionnaires. This consumption measure covers food, housing and other non-food items, and includes imputations for consumption from sources other than market purchases. These imputations include consumption from the output of own production (mostly agriculture, but also from non-farm enterprises), wage payments and transfers received in kind, and imputed rent from owner-occupied dwellings. An imputation is also made for consumption services derived from durable consumer goods owned by the household, rather than including expenditure on the acquisition of such goods (these are lumpy expenditures, e.g. purchasing a car, more like investment rather than consumption).
Total consumption expenditure is estimated for a twelve-month period based on information collected with the questionnaire. In the case of frequent purchases (e.g. food purchases, consumption of own produced food, frequently purchased non-food items such as soap, tobacco) this is estimated by grossing up responses relating to a shorter recall period. Households received multiple visits at regular intervals of a few days in the course of the survey (in GLSS 3 eight visits at two-day intervals in rural areas and eleven visits at three-day intervals in urban areas; seven visits at 5-day interval in the case of GLSS 4; and 11 visits at three days interval in GLSS 5). In each case, in all but the first two visits, they were asked about their purchases of each item since the last visit, and the answers to these “bounded recall” questions (recall relative to a fixed reference point) was used as the basis for estimating annual expenditure or consumption. Similar principles were used to estimate annual expenditure on frequently purchased non-food items and on consumption of own produced food (valuing items at the price at which they could have been sold). In the case of consumption of own produced food, allowance was made for the number of months in which an item was normally consumed.
The recall period for frequently purchased or consumed items did change between GLSS 3, GLSS 4 and GLSS 5, and experimental evidence for Ghana and elsewhere suggest that lengthening the recall period causes respondents to progressively forget more items of expenditure. A study for Ghana by Scott and Amenuvegbe (1990) found that, on average, respondents forgot 2.9 percent of expenditure for each day by which the recall period was lengthened (up to seven days). Given this evidence, this figure was used to estimate what each household’s expenditure on frequent purchases in GLSS 3 would have been had the same recall period been used as for GLSS 4 and GLSS 5.
A longer recall period, generally three or twelve months, was used in collecting information on less frequently purchased consumption items (e.g. clothing and footwear); this again is grossed up as necessary. As noted above, purchases of durable goods were not included in this, and some other expenditure items deemed not to be associated with increases in welfare were also excluded such as expenditure on hospital stays. This is also a lumpy item, and it would not be reasonable to regard a household as being significantly better off because it had to make a large expenditure on an emergency operation, say. Everyday medical expenses were though included in the consumption measure.
In the case of owner occupied dwellings, imputed rents were estimated based on a hedonic equation, which related rents of rented housing to characteristics, and uses this to estimate rental values for owner-occupied dwellings based on their characteristics and amenities. Consumption flows (use values) for durable goods were estimated based on assumed depreciation rates. In both cases the procedures used for GLSS 3, GLSS 4 and GLSS 5 were identical.
The remaining items in the estimate of household consumption relate to the value of wage payments received in kind, and consumption of the output of non-farm enterprises owned and operated by the household. The sum of all the items in Table A2.1 gives the estimate of total household consumption expenditure, which is expressed in nominal values (current prices).
Table A2.1: Estimation of total household consumption expenditure from the GLSS 3, GLSS 4, and GLSS5 surveys Element of total household
consumption
Composition Source of data in GLSS
questionnaire
Notes
Expenditure on food, beverages and tobacco
Expenditure on about 120 commodities (based on pattern in several short recall periods in the past month)
Section 9B Consumption of food commodities from own production, valued by
respondents at prices at which they could be sold
Section 8H Consumption of own produced
food
Wage income received in form of food (based on payment interval reported by respondents)
Section 4 Expenditure on frequently purchased non-food items (based on
pattern in several short recall periods in the past month)
Section 9A2 Section 9B in GLSS5 Expenditure on less-frequently purchased non-food goods and
services (based on pattern over last 3 or last 12 months)
Section 9A1 Excluding purchases of durable goods and expenditure on hospital stays
Expenditure on education (based on pattern for each child in past 12 months)
Section 2 Expenditure on non-food
items
Expenditure on household utilities: water, electricity, garbage disposal (based on payment interval reported by respondents)
Section 7 Actual rental expenditure (based on payment interval reported by
respondents)
Section 7
Imputed rent of owner occupied dwellings Section 7 Estimated based on hedonic regression equation
Expenditure on housing
Wage income received as subsidized housing (based on payment interval reported by respondents)
Section 4 Durable goods user values Section 12B Consumption from output of non-farm enterprises (based on two
week period)
Section 10D Imputed expenditure on non-
food items
Wage income in kind in forms other than food and housing (based on payment interval reported by respondents)
Allowing for differences in the size and composition of households
Adjustments are needed to construct a standard of living measure that takes into account differences in the size and composition of households. A simple way of doing this would be to divide total consumption by household size to obtain consumption expenditure per capita. But this would not allow for the fact that different members (e.g. young children and adults) are likely to have different consumption needs. To account for differences in needs, the idea is to measure household size in equivalent adults, using an appropriate adult equivalence scale that reflects the relative consumption needs of different household members (e.g. based on age, gender). The equivalence scale used is based on calorie requirements commonly used in nutritional studies in Ghana, as provided in Table A2.1. Calorie requirements are distinguished by age category and gender, information which is also reported in the household questionnaire. This information is used to estimate household size in number of adult equivalents.
Table A2.1: Recommended energy intakes per person according to gender and age
Category Age (years) Average energy allowance per day (kcal)
Equivalence scale Infants 0 - 0.5 650 0.22 0.5 - 1.0 850 0.29 Children 1 – 3 1300 0.45 4 – 6 1800 0.62 7 – 10 2000 0.69 Males 11 – 14 2500 0.86 15 – 18 3000 1.03 19 – 25 2900 1.00 25 - 50 2900 1.00 51+ 2300 0.79 Females 11 - 14 2200 0.76 15 - 18 2200 0.76 19 - 25 2200 0.76 25 - 50 2200 0.76 51+ 1900 0.66 Source: Recommended Dietary Allowances, 10th edition, (Washington D.C.: National Academy Press, 1989).
The standard of living measure is then measured by dividing the estimate of total household consumption expenditure in constant prices by household size measured in number of equivalent adults. The poverty analysis is based on the distribution of this standard of living measure over all households in the sample, weighting each household by its size in number of persons. This household size weight means that for example a poor household of six members is given twice the weight of an equally poor household of three persons. Each individual (rather than each household) in the sample is given equal weight. Note that this equal weighting of all individuals when estimating poverty measures violates the assumption that individuals differ in needs, but this is still what is done in practice in empirical studies on poverty.
Allowing for cost of living variations
Having estimated total household consumption expenditure, further steps are needed before it is possible to compare standards of living across households. Because the standard of living is expressed in nominal terms, it must be adjusted to allow for variations in prices faced by households. Three sources of variation are relevant for purposes of this study: (i) differences in the cost of living between different localities at a point in time; (ii) variations in prices within the time periods covered by the surveys, which can occur due to inflation, seasonality and other reasons; (iii) most importantly (in comparing trends between the three GLSS rounds) inflation between the GLSS 3, GLSS 4 and GLSS 5 (substantial in this case).
A cost of living index was constructed capturing these different dimensions of variation. Geographic differences in the cost of living were estimated based on the GLSS 4 price questionnaire, in conjunction with expenditure data from the GLSS 4 household questionnaire. Based on five localities, Paasche cost of living indices were constructed for food and non-food separately. The hedonic regression equation was used to estimate a housing cost of living index by comparing rental values for a dwelling with the same characteristics and amenities in each locality. These procedures give the geographic cost of living indices reported in Table A2.2. The regional cost of living index based on GLSS 4 presented in Table A2.2 indicates that there are significant differences in the prices of food and housing, with urban areas in general and Accra in particular being more expensive for these items than rural areas. The prices of other non-food items are much more uniform. The regional cost of living index is a weighted average of these three regional sub-indices.
Table A2.2: Regional cost of living indices
Food index Non food index Housing index
Accra 1.0000 1.0000 1.0000
Other Urban 0.9183 0.9086 0.6442 Rural Coastal 0.8832 0.9753 0.6149 Rural Forest 0.8212 0.9839 0.5296 Rural Savannah 0.7310 1.0484 0.4491 Source: Computed from the Ghana Living Standards Survey, 1998/99.
Variations in prices within and between the sample years were then allowed by using the Consumer Price Index, using separate series for food and non-food, as well as for Accra, other urban and rural areas. A single overall cost of living index was constructed combining the geographic and over time variations. This was used to deflate the estimate of total household consumption expenditure, so that it was now expressed in the constant prices of a reference locality and time period (Accra in January 2006).
Construction of the poverty lines
The approach taken was to anchor the poverty lines in calorie requirements. The method involves examining the average consumption basket of the bottom x percent (say 50 percent) of the population ranked by the standard of living measure, and computing how many calories this
basket provides per adult equivalent. The quantities of each item consumed in the basket can then be scaled up (or down) in the appropriate proportion to compute the basket with this composition, which would provide the minimum calorie requirements (2900 kilocalories per equivalent adult based on the scale used in Ghana). This provides an estimate of the food expenditure required to attain 2900 kilocalories, based on the consumption basket of the poorest x percent of the distribution. Obviously, one of the issues is the choice of x. It is worth noting that some observers find 2900 Kcal too high given that most poverty profiles in other developing countries use between 2100 and 2400 Kcal for their poverty lines. Yet those countries usually construct a per capita welfare measure while ours is based on equivalent adult. It would be easy to show that our level of kilocalories on a per capita basis would be 2202 kcal per day.
Taking account of non-food needs is more difficult. Following common practice in other developing countries (Ravallion, 1994), the non-food poverty line is based on the expenditure devoted to non-food items of those households whose total consumption expenditure is at the level of (or close to) the food poverty line. This is based on the principle that these non-food consumption items are essential for households, so that they will even forgo meeting their calorie requirements (or consume an “inferior” basket) in order to purchase them.
This poverty line methodology had been used in the previous poverty profile based on GLSS 3 and 4 (GSS, 2000). The methodology used suggests food poverty line of, in round figures, 700,000 when x=50 percent (slightly lower for lower values of x), while allowing for non-food requirements suggests an overall poverty line of approximately 900,000 cedis per equivalent adult per year in Accra, January 1999 prices. As shown in World Bank (1995), this line represents roughly $1 a day. This latter line would be used as the overall poverty line for Ghana. The lower poverty line of 700,000 is used as an extreme poverty line; people whose standard of living measure lies below this would not be able to meet their calorie requirements even if they spent their entire budget on food.
These same poverty lines of 700,000 and 900,000 cedis were used for the analysis of the 1991/92 and 2005/2006 surveys but they were inflated using locality specific Consumer Price Index (CPI) provided by GSS, backward for the 1991/92 survey and forward to January 2006 prices for the 2005/2006 survey, yielding extreme and overall poverty lines of 2,884,700 cedis and 3,708,900 cedis in 2006. Those lines take into account price differentials between the different localities. In 2006 in local prices the higher line can be translated to 3,708,900 (Accra); 2,773,170 (Other Urban); 3,146,220 (Rural Coastal); 3,034,800 (Rural Forest) and 2,850,120 (Rural Savannah).
Standard errors of poverty measures
As any other statistics computed from survey data, poverty measures have standard errors, and these standard errors must be considered when assessing whether changes in poverty over time, or differences in poverty between groups can be considered as statistically significant. To complement the estimates provided in the main text, this section provides the standard errors of the main poverty measures estimated with the three GLSS surveys.
Annex Table 1: Poverty measures by urban/rural location with standard errors and confidence intervals
2006 1998/99 1991/92
Poverty Std. Err. [95% Conf. Int.] Poverty Std. Err. [95% Conf. Interval] Poverty Std. Err. [95% Conf. Interval]
Headcount index Urban 10.8 1.4 8.1 13.6 19.4 2.6 14.2 24.6 27.7 2.3 23.2 32.3 Rural 39.2 2.0 35.3 43.1 49.6 2.6 44.4 54.8 63.6 1.6 60.4 66.8 Ghana 28.5 1.5 25.6 31.5 39.5 2.3 35.0 43.9 51.7 1.7 48.4 55.0 Poverty gap Urban 3.1 0.5 2.0 4.1 5.3 0.8 3.8 6.9 7.4 1.0 5.4 9.4 Rural 13.5 0.9 11.7 15.3 18.2 1.6 15.1 21.3 24.0 1.0 22.0 26.0 Ghana 9.6 0.6 8.3 10.9 13.9 1.2 11.6 16.2 18.5 0.9 16.8 20.2
Squared poverty gap
Urban 1.3 0.3 0.8 1.8 2.1 0.3 1.4 2.7 2.9 0.6 1.7 4.1 Rural 6.6 0.6 5.5 7.7 8.9 1.0 7.0 10.9 11.7 0.7 10.4 13.0 Ghana 4.6 0.4 3.8 5.3 6.6 0.7 5.2 8.0 8.8 0.5 7.7 9.8 Source: Authors.
Annex Table 2: Poverty measures by region with standard errors and confidence intervals
2006 1998/99 1991/92
Poverty Std. Err. [95% Conf. Int.] Poverty Std. Err. [95% Conf. Interval] Poverty Std. Err. [95% Conf. Interval]
Headcount index Western 18.4 3.3 11.9 24.9 27.3 3.6 20.1 34.4 59.6 3.7 52.3 66.8 Central 19.9 3.5 13.1 26.8 48.4 3.6 41.3 55.6 44.3 4.0 36.5 52.0 Greater Accra 11.8 2.5 6.9 16.7 5.2 1.6 2.1 8.3 25.8 3.4 19.2 32.4 Volta 31.4 4.4 22.8 40.1 43.7 5.2 33.5 53.8 48.0 4.1 39.9 56.1 Eastern 15.1 2.7 9.7 20.4 37.7 2.9 32.0 43.5 57.0 4.6 48.0 66.0 Ashanti 20.3 2.5 15.3 25.3 27.7 4.8 18.2 37.3 41.2 3.7 33.9 48.5 Brong Ahafo 29.5 4.0 21.6 37.3 35.8 5.8 24.5 47.2 65.0 4.9 55.3 74.7 Northern 52.3 5.9 40.8 63.9 69.2 6.4 56.7 81.7 63.4 6.6 50.5 76.4 Upper East 70.4 4.8 61.0 79.7 83.9 9.2 65.8 102.0 88.4 3.2 82.2 94.6 Upper West 87.9 3.3 81.3 94.4 88.2 5.6 77.1 99.2 66.9 5.0 57.1 76.7 Ghana 28.5 1.5 25.6 31.5 39.5 2.3 35.0 43.9 51.7 1.7 48.4 55.0 Poverty gap Western 4.2 1.0 2.2 6.3 7.0 1.0 5.0 8.9 20.5 1.9 16.7 24.3 Central 4.3 1.0 2.4 6.2 14.8 1.6 11.7 17.9 12.9 2.2 8.6 17.2 Greater Accra 3.1 0.8 1.5 4.6 1.1 0.3 0.4 1.7 6.3 1.1 4.1 8.5 Volta 7.3 1.2 4.9 9.6 15.6 2.4 11.0 20.3 15.9 1.6 12.7 19.1 Eastern 3.3 0.8 1.7 4.9 9.9 1.3 7.3 12.4 20.1 2.3 15.5 24.6 Ashanti 5.2 0.8 3.7 6.7 8.5 1.9 4.8 12.1 12.9 1.5 9.9 15.9 Brong Ahafo 7.8 1.4 5.1 10.5 9.8 2.0 5.8 13.7 22.8 2.4 18.1 27.4 Northern 20.7 2.7 15.4 26.1 29.9 4.1 21.8 38.1 29.9 4.2 21.6 38.1 Upper East 32.7 3.5 26.0 39.5 38.8 11.8 15.6 62.0 41.3 4.7 32.1 50.5 Upper West 48.0 4.1 39.9 56.0 44.0 5.0 34.2 53.8 28.7 3.2 22.3 35.0 Ghana 9.6 0.6 8.3 10.9 13.9 1.2 11.6 16.2 18.5 0.9 16.8 20.2
Squared poverty gap
Western 1.4 0.5 0.5 2.4 2.5 0.4 1.7 3.2 9.1 1.1 7.0 11.2 Central 1.4 0.3 0.7 2.0 6.0 0.8 4.4 7.6 5.7 1.4 3.0 8.4 Greater Accra 1.1 0.3 0.5 1.7 0.3 0.1 0.1 0.5 2.3 0.5 1.3 3.4 Volta 2.4 0.5 1.6 3.3 7.4 1.3 4.8 10.0 6.6 0.8 5.0 8.1 Eastern 1.3 0.4 0.6 1.9 3.8 0.7 2.5 5.2 9.1 1.3 6.5 11.6 Ashanti 1.9 0.3 1.3 2.5 3.7 1.0 1.7 5.7 5.6 0.8 4.0 7.1 Brong Ahafo 3.0 0.7 1.6 4.4 3.9 1.1 1.7 6.2 10.2 1.3 7.7 12.7 Northern 10.5 1.6 7.4 13.6 15.5 2.7 10.1 20.8 17.2 2.9 11.4 23.0 Upper East 18.4 2.5 13.4 23.4 22.7 8.4 6.3 39.2 23.3 3.4 16.6 30.1 Upper West 30.2 3.4 23.6 36.9 25.1 3.9 17.5 32.7 15.2 2.2 11.0 19.5 Ghana 4.6 0.4 3.8 5.3 6.6 0.7 5.2 8.0 8.8 0.5 7.7 9.8 Source: Authors.
ANNEX 3